Across 22 brand-SKUs in Beauty, Financial Services, and Travel, scoring 592 specific brand facts: 75.7% (95% CI 72.1–79.0%) of the facts a large language model stated it possessed about a brand were not deployed when the model made an actual purchase recommendation. At scale across 1,427 brand probes spanning ten industries: 87.3% (95% CI 85.5–88.9%) of brands explicitly anchored at the first turn of a conversation are displaced by a competitor by the fourth turn. We call this the Linkage Gap - the structural divide between what AI systems possess about a brand and what they deploy at the moment of recommendation. The gap is systematic, brand-asymmetric, and not closed by possession-side investments (knowledge graphs, training-time partnerships) or Layer 2 investments (NLWeb adoption, citation engineering, AI-readable site standards). It is closed when the right brand fact is surfaced at the conversational moment of decision - a result confirmed at 100% deployment rate and 80% brand-recommendation conversion in controlled counterfactual testing. This paper introduces the Linkage Gap, presents the empirical evidence, explains the mechanism through a two-regime model of foundation-model behaviour, integrates three independent academic confirmations, and sets out the architectural argument for a corrected three-axis framework - possession / Layer 2 Mention / Layer 3 Activation - as the missing investment category in AI brand strategy. Intended for chief marketing officers, heads of digital, AI strategy leads, and the agency teams that serve them.
Sheals et al. (Thu,) studied this question.